Towards Robust Prompts on Vision-Language Models
Paper
ā¢
2304.08479
ā¢
Published
Ahmad Beirami
beirami
AI & ML interests
None yet
Recent Activity
authored
a paper
16 days ago
Towards Robust Prompts on Vision-Language Models
authored
a paper
16 days ago
Enhancing Group Fairness in Online Settings Using Oblique Decision
Forests
authored
a paper
16 days ago
Situated and Interactive Multimodal Conversations
Organizations
beirami's activity
authored
7
papers
16 days ago
Enhancing Group Fairness in Online Settings Using Oblique Decision Forests
Paper
ā¢
2310.11401
ā¢
Published
Situated and Interactive Multimodal Conversations
Paper
ā¢
2006.01460
ā¢
Published
A Systematic Survey of Prompt Engineering on Vision-Language Foundation Models
Paper
ā¢
2307.12980
ā¢
Published
ā¢
1
Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment
Paper
ā¢
2411.18688
ā¢
Published
InfAlign: Inference-aware language model alignment
Paper
ā¢
2412.19792
ā¢
Published
ā¢
1
Data-augmented phrase-level alignment for mitigating object hallucination
Paper
ā¢
2405.18654
ā¢
Published
authored
a
paper
about 1 month ago
authored
a
paper
8 months ago
authored
a
paper
11 months ago
reacted to
gsarti's
post with ā¤ļø
about 1 year ago
Post
š Today's pick in Interpretability & Analysis of LMs: Gradient-Based Language Model Red Teaming by N. Wichers, C. Denison and
@beirami
This work proposes Gradient-Based Red Teaming (GBRT), a red teaming method for automatically generating diverse prompts inducing an LM to output unsafe responses.
In practice, prompts are learned by scoring LM responses with a safety-trained probing classifier, and back-propagating through frozen classifier and LM to update the prompt.
Authors experiment with variants of GBRT aimed at inducing realistic prompts in an efficient way, and GBRT prompts are more likely to generate unsafe responses than those found by established RL-based red teaming methods. Moreover, these attacks are shown to succeed even when the LM has been fine-tuned to produce safer outputs.
š Paper: In-Context Language Learning: Architectures and Algorithms (2401.12973)
š» Code: https://github.com/google-research/google-research/tree/master/gbrt
This work proposes Gradient-Based Red Teaming (GBRT), a red teaming method for automatically generating diverse prompts inducing an LM to output unsafe responses.
In practice, prompts are learned by scoring LM responses with a safety-trained probing classifier, and back-propagating through frozen classifier and LM to update the prompt.
Authors experiment with variants of GBRT aimed at inducing realistic prompts in an efficient way, and GBRT prompts are more likely to generate unsafe responses than those found by established RL-based red teaming methods. Moreover, these attacks are shown to succeed even when the LM has been fine-tuned to produce safer outputs.
š Paper: In-Context Language Learning: Architectures and Algorithms (2401.12973)
š» Code: https://github.com/google-research/google-research/tree/master/gbrt
authored
2
papers
about 1 year ago
authored
a
paper
over 1 year ago